Multi-modal large model-based reimbursement receipt auditing method and device and electronic equipment

Abstract

The invention provides a reimbursement receipt auditing method and device based on a multi-modal large model and electronic equipment, and relates to the technical field of artificial intelligence. The method comprises the following steps: acquiring auditing request information about reimbursement receipts input by a user; inputting the auditing request information into a multi-modal large model to obtain a tool calling information sequence corresponding to the reimbursement document auditing process; wherein the multi-modal large model is used for performing intention recognition based on the auditing request information, and determining a tool calling information sequence based on an intention recognition result and preset tool information; according to the tool calling information sequence, corresponding tools are called in sequence to execute operations, and contents returned by all the calling operations are obtained; and inputting content returned based on each calling operation into the multi-modal large model to obtain an auditing result of the reimbursement document. The reimbursement document auditing process can be automatically arranged based on the large model, the flexibility of the auditing process can be greatly improved, and the labor cost can be saved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a reimbursement document review method, device and electronic equipment based on a multimodal large model. Background Art

[0002] In related technologies, in the process of processing financial reimbursement documents, RPA (Robotic Process Automation) is used to configure relevant audit rules to complete financial audit related work. Although this solution has high processing efficiency, it has poor flexibility. The audit process usually involves many rules. When the enterprise management regulations change, it is difficult to maintain the RPA audit rules. Summary of the invention

[0003] The embodiments of the present invention provide a reimbursement document review method, device and electronic device based on a multimodal large model, which are used to solve the technical problem of poor flexibility in the prior art when performing audit work based on RPA robots in scenarios where audit rules change.

[0004] In a first aspect, an embodiment of the present invention provides a reimbursement document review method based on a multimodal big model, comprising: obtaining review request information about the reimbursement document input by a user; inputting the review request information into the multimodal big model to obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; according to the tool call information sequence, calling the corresponding tools in turn to perform operations to obtain the content returned by each call operation; based on the content returned by each call operation, inputting it into the multimodal big model to obtain the review result of the reimbursement document.

[0005] In some embodiments, the step of inputting the review request information into the multimodal big model to obtain a tool call information sequence corresponding to the reimbursement document review process includes: based on a retrieval enhancement generation algorithm, matching and retrieving the review request information with a reimbursement document knowledge base to obtain target knowledge information; inputting the review request information and the target knowledge information into the multimodal big model to obtain the tool call information sequence.

[0006] In some embodiments, the step of inputting the review request information into the multimodal big model to obtain a tool call information sequence corresponding to the reimbursement document review process includes: inputting the review request information into the multimodal big model to obtain missing parameter information; initiating a user dialogue based on the missing parameter information; receiving user reply information, and inputting the user reply information and the review request information into the multimodal big model together to obtain the tool call information sequence.

[0007] In some embodiments, the tool call information sequence includes API interface call information; calling the corresponding tools in sequence to perform operations according to the tool call information sequence to obtain the content returned by each call operation, including: determining the target API interface and target parameter information according to the API interface call information; calling the target API interface based on the target parameter information to obtain the content returned by the target API interface.

[0008] In some embodiments, the tool call information sequence includes database query statement information; calling the corresponding tools in sequence to perform operations according to the tool call information sequence, and obtaining the content returned by each call operation includes: executing the database query statement information and obtaining the content returned by the database query operation.

[0009] In some embodiments, the tool call information sequence includes task script information; calling corresponding tools in sequence according to the tool call information sequence to obtain the content returned by each call operation includes: executing the task script information based on a script parser to obtain the content returned by the script execution operation.

[0010] In a second aspect, an embodiment of the present invention provides a reimbursement document review device based on a multimodal big model, comprising: a first acquisition module, used to obtain review request information about the reimbursement document input by a user; a second acquisition module, used to input the review request information into the multimodal big model, and obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; a calling module, used to call the corresponding tool to perform operations in sequence according to the tool call information sequence, and obtain the content returned by each calling operation; a third acquisition module, used to input the content returned by each calling operation into the multimodal big model, and obtain the review result of the reimbursement document.

[0011] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor and a memory storing a computer program, wherein when the processor executes the program, the multimodal large model reimbursement document review method described in the first aspect is implemented.

[0012] In a fourth aspect, an embodiment of the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the multimodal large model reimbursement document review method described in the first aspect is implemented.

[0013] In a fifth aspect, an embodiment of the present invention provides a computer program product, including a computer program, which, when executed by a processor, implements the multimodal large model reimbursement document review method described in the first aspect.

[0014] The reimbursement document review method, device and electronic device based on the multimodal big model provided by the embodiment of the present invention obtain the review request information about the reimbursement document input by the user; input the review request information into the multimodal big model to obtain the tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and the preset tool information; according to the tool call information sequence, the corresponding tools are called in sequence to perform operations to obtain the content returned by each call operation; based on the content returned by each call operation, the content is input into the multimodal big model to obtain the review result of the reimbursement document. The present invention can realize the automatic arrangement of the reimbursement document review process based on the big model, which is not affected by the changes in the enterprise review rules, which can not only greatly improve the flexibility of the review process, but also save labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 One of the flow diagrams of the reimbursement document review method based on the multimodal large model provided in an embodiment of the present invention; Figure 2 A second flow chart of a method for reviewing reimbursement documents based on a multimodal large model provided in an embodiment of the present invention; Figure 3 A schematic diagram of the architecture of a method for reviewing reimbursement documents based on a multimodal large model provided in an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a reimbursement document review device based on a multimodal large model provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] The present invention provides a method, device and electronic equipment for reviewing reimbursement documents based on a multimodal large model.

[0019] Figure 1 This is a flow chart of a method for reviewing reimbursement documents based on a multimodal large model provided by an embodiment of the present invention. Figure 1 As shown, the method may include the following steps.

[0020] Step 101, obtaining the review request information about the reimbursement document input by the user.

[0021] In some embodiments, a user can input review request information about a reimbursement document through a user interaction interface, and the user can be a user corresponding to any approval node in the reimbursement document approval flow. It can be understood that the users corresponding to each node in the approval flow may have different concerns about the reimbursement document, so the user can initiate a review request information about the reimbursement document based on his or her own approval needs.

[0022] In some embodiments, the review request information of the reimbursement document may include natural language information input by the user and identification information of the reimbursement document. The natural language information may be voice information input by the user or text description information input by the user. The natural language information input by the user may be pre-set, such as pre-setting the natural language information corresponding to the approval requirements, and when the approval flow node reaches the user, the approval request information is automatically initiated. The natural language information input by the user may also be input in real time, such as when the user opens the approval information in the interactive interface, after triggering the operation of inputting the review requirements, the voice information may be input, and the review request information may be automatically initiated after confirming the input voice information.

[0023] It should be noted that the executor of the embodiment of the present invention may be an agent, that is, after receiving the approval request information of the reimbursement document initiated by the user, the agent automatically approves the reimbursement document and feeds back the approval result of the document to the user through the user interface.

[0024] Step 102, input the review request information into the multimodal big model to obtain the tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and the preset tool information.

[0025] The multimodal large model can be a general natural language large model, such as the Qwen-72B large model. In some embodiments, the multimodal large model has learned the ability to recognize intent, the mapping relationship between intent recognition results and tool calls, and the calling methods of each tool, as well as the ability to output audit results based on the return content of the tool call.

[0026] In some embodiments, training can be performed based on the pre-trained Qwen-72B large model to obtain a multimodal large model. The training samples in the training process may include: natural language information samples, calling tool information labels, and description information samples of each calling tool.

[0027] In some embodiments, the calling tools supported by the multimodal big model may include API interfaces, database query tools, RPA robots, script parsers, etc.

[0028] In some embodiments, the tool calling information sequence may include multiple tool calling information, such as which tools need to be called, how to call each tool, what information needs to be obtained, etc. The calling of each tool may have a sequence or may not consider the sequence. Whether there is a sequence can be marked in the tool calling information sequence.

[0029] As an example, the tool call information sequence may include API interface call information, such as the API interface to be called and parameter information. The tool call information sequence may also include database query statement information, such as SQL statement information for querying the corresponding database. The tool call information sequence may also include script information.

[0030] In some embodiments, the multimodal big model can determine whether there are missing parameters when generating the interface call information sequence. If there are missing parameters, the missing parameter information can be output, and the agent can initiate a dialogue with the user based on the missing parameter information, and guide the user to accurately describe the business scenario through multiple rounds of dialogue to supplement the missing parameters.

[0031] As a possible implementation method, the implementation process of inputting the review request information into the multimodal big model to obtain the tool call information sequence corresponding to the reimbursement document review process may include: inputting the review request information into the multimodal big model to obtain missing parameter information; initiating a user dialogue based on the missing parameter information; receiving user reply information, and inputting the user reply information and the review request information into the multimodal big model together to obtain a tool call information sequence.

[0032] Step 103, according to the tool calling information sequence and the document information of the reimbursement document, call the corresponding tools in sequence to perform operations, and obtain the content returned by each calling operation.

[0033] In some embodiments, there is no order between each tool call in the tool call information sequence, and the corresponding tool can be called in parallel according to the tool call information sequence to perform operations and obtain the content returned by each call operation.

[0034] In some embodiments, there is a sequence between each tool call in the tool call information sequence. For example, to review whether the reimbursement amount exceeds the project budget, it is necessary to first query the project information and reimbursement amount of the reimbursement document, and then query the project budget based on the project information. It is necessary to supplement the parameters required for the next tool call based on the return content of the previous tool call, and then execute the next tool call.

[0035] In some embodiments, the tool call information sequence includes API interface call information, and then the corresponding tool is called in sequence to perform operations according to the tool call information sequence and the document information of the reimbursement document, and the execution process of obtaining the content returned by each call operation includes: determining the target API interface and target parameter information according to the API interface call information; calling the target API interface based on the target parameter information to obtain the content returned by the target API interface. If the API interface call information has a sequential correlation with the previous tool call, it is necessary to determine the target parameter information based on the content returned by the previous tool call.

[0036] In some embodiments, the tool call information sequence may also include the call information of the RPA robot. Based on the call information of the RPA robot, the relevant API interface of the RPA robot may be called to perform corresponding information query and obtain the return content. For example, to review the invoice issuing unit, the RPA robot may be called to query whether the invoice issuing unit is a defaulting supplier, etc., to identify whether there is any violation in the invoice.

[0037] As an example, for the API interface call information for verifying the authenticity of invoices, the tool call process is equivalent to calling the invoice verification API provided by the invoice verification platform to verify the authenticity of the invoice. Among them, the attachment information of the invoice can be obtained based on the API interface of the financial reimbursement system.

[0038] In some embodiments, the tool call information sequence includes database query statement information. According to the tool call information sequence, the corresponding tools are called in sequence to perform operations, and the contents returned by each call operation are obtained, including: executing database query statement information, and obtaining the contents returned by the database query operation. For example, by querying the budget management system, the financial budget surplus funds of the corresponding project are obtained. Among them, the database query statement can be a corresponding SQL statement. If the parameters involved in the query statement include the return content of other tool calls, it is necessary to supplement the parameters into the database query statement in combination with the contents returned by the previous tool call operation.

[0039] In some embodiments, the tool call information sequence also includes task script information. According to the tool call information sequence, the corresponding tools are called in sequence to perform operations, and the content returned by each call operation is obtained, including: executing the task script information based on the script parser, and obtaining the content returned by the script execution operation. For example, in the process of classifying the attachments uploaded by the user, by generating a call code, calling a preset attachment classification algorithm, completing the attachment category parsing and returning the result.

[0040] Step 104, based on the content returned by each calling operation, input it into the multimodal big model to obtain the review result of the reimbursement document.

[0041] According to the reimbursement document review method based on the multimodal big model of the embodiment of the present invention, the review request information about the reimbursement document input by the user is obtained; the review request information is input into the multimodal big model to obtain the tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and the preset tool information; according to the tool call information sequence, the corresponding tools are called in sequence to perform operations to obtain the content returned by each call operation; based on the content returned by each call operation, the content is input into the multimodal big model to obtain the review result of the reimbursement document. The present invention can realize the automatic arrangement of the reimbursement document review process based on the big model, which is not affected by the changes in the enterprise review rules, which can not only greatly improve the flexibility of the review process, but also save labor costs.

[0042] Figure 2 The second flowchart of the reimbursement document review method based on the multi-modal large model provided by the embodiment of the present invention. Figure 2 As shown, based on the above embodiments, Figure 1The implementation process of step 102 in may include the following steps.

[0043] Step 201, based on the retrieval enhancement generation algorithm, the review request information and the reimbursement document knowledge base are matched and searched to obtain the target knowledge information.

[0044] In some embodiments, a reimbursement document knowledge base may be pre-built, for example, it may include a budget management knowledge base, a cash flow management knowledge base, a cost control knowledge base, an asset management knowledge base, a financial data knowledge base, and a human resources knowledge base, etc. Each knowledge base may include information such as a knowledge base name, a knowledge base summary, and knowledge base data, wherein the knowledge base summary describes the main content and applicable scenarios in the knowledge base to improve the query hit rate of retrieval enhancement.

[0045] In some embodiments, in order to improve the accuracy of the information sequence called by the large model output tool, the pre-built reimbursement document knowledge base can be matched with the audit request information to obtain relevant knowledge content. The reimbursement document knowledge base can be a knowledge base that divides and vectorizes relevant information, vectorizes the audit request information, and matches the vector expression of the audit request information with the content in the reimbursement document knowledge base to obtain relevant target knowledge information.

[0046] In some embodiments, the usage scenarios of the reimbursement document knowledge base for each segmented scenario can be used as background knowledge and automatically superimposed in the prompt words. After the user enters the review request information, the user is guided through multiple rounds of dialogue to accurately describe the business scenarios they are concerned about, so as to match the corresponding content in the knowledge base through a chain of thought.

[0047] Step 202, input the audit request information and target knowledge information into the multimodal big model to obtain a tool call information sequence.

[0048] In other words, the multimodal large model combines request information and target knowledge information to determine the user's intention, which tools need to be called during the audit process, which data needs to be obtained through tool calls, and how to call the tools, in order to generate the corresponding tool call information sequence.

[0049] According to the reimbursement document review method based on the multimodal big model of the embodiment of the present invention, based on the retrieval enhancement generation algorithm, the review request information and the reimbursement document knowledge base are matched and retrieved to obtain the target knowledge information, and the request information and the target knowledge information are input into the multimodal big model to obtain the tool call information sequence. Through the retrieval enhancement generation algorithm, the accuracy of the tool call information sequence generated by the big model can be improved, thereby improving the efficiency and accuracy of the reimbursement document approval.

[0050] Figure 3FIG. 1 is a schematic diagram of the architecture of a method for reviewing reimbursement documents based on a multimodal large model according to an embodiment of the present invention. Figure 3 As shown in the figure, the architecture can include the model layer, tool layer, scheduling layer and business layer. Among them, the model layer contains the artificial intelligence algorithm models that need to be called, such as: multimodal large model, image classification algorithm, optical character recognition algorithm, etc. The tool layer mainly includes the tools required to complete the reimbursement of reimbursement documents, such as: code parsing module, API call module, RPA call module and database query module. The scheduling layer is used to receive user input and complete the automatic audit of financial documents by calling various models and tools. It mainly includes workflow orchestration module and retrieval enhancement generation module. The business layer mainly connects to various external business systems, such as: financial reimbursement system, budget management system, negative list system and invoice verification platform.

[0051] The workflow orchestration module mainly receives natural language instructions input by users, uses pre-trained large models to complete intent recognition, analyzes the external capabilities required to complete the user-specified work, analyzes the results returned by each link after completing the capability call, and finally outputs the execution results. The types of capabilities it supports include but are not limited to: API interface, database query, RPA robot, and script parser.

[0052] The search enhancement module includes multiple sub-domain knowledge bases, such as the project budget management knowledge base, cash flow management knowledge base, cost control knowledge base, asset management knowledge base, financial data knowledge base, and human resources knowledge base. Each knowledge base contains information such as the knowledge base name, knowledge base summary, and knowledge base data. The main content and applicable scenarios of the knowledge base are described in the knowledge base summary to improve the query hit rate of large-model knowledge retrieval enhancement. The applicable scenarios of the financial knowledge base for each sub-scenario are used as background knowledge and automatically superimposed in the prompt words. When users ask questions, they are guided to accurately describe the business scenarios they are concerned about through multiple rounds of dialogue, making it easier to match the corresponding knowledge base through a chain of thoughts.

[0053] Figure 4 The schematic diagram of the structure of the reimbursement document review device based on the multi-modal large model provided by the embodiment of the present invention. Figure 4As shown, the device includes: a first acquisition module 410, a second acquisition module 420, a calling module 430 and a third acquisition module 440. The first acquisition module 410 is used to obtain the review request information about the reimbursement document input by the user; the second acquisition module 420 is used to input the review request information into the multimodal large model to obtain the tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal large model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and the preset tool information; the calling module 430 is used to call the corresponding tool to perform operations in sequence according to the tool call information sequence, and obtain the content returned by each call operation; the third acquisition module 440 is used to input the content returned by each call operation into the multimodal large model to obtain the review result of the reimbursement document.

[0054] In some embodiments, the second acquisition module 420 is specifically used to: based on the retrieval enhancement generation algorithm, match and search the review request information with the reimbursement document knowledge base to obtain the target knowledge information; input the review request information and the target knowledge information into the multimodal large model to obtain a tool call information sequence.

[0055] In some embodiments, the second acquisition module 420 is also used to: input the review request information into the multimodal big model to obtain missing parameter information; initiate a user dialogue based on the missing parameter information; receive user reply information, input the user reply information and the review request information into the multimodal big model together, and obtain a tool call information sequence.

[0056] In some embodiments, the tool call information sequence includes API interface call information; the call module 430 is specifically used to: determine the target API interface and target parameter information based on the API interface call information; call the target API interface based on the target parameter information to obtain the content returned by the target API interface.

[0057] In some embodiments, the tool call information sequence includes database query statement information; the call module 430 is further used to: execute the database query statement information, and obtain the content returned by the database query operation.

[0058] In some embodiments, the tool call information sequence includes task script information; the call module 430 is further used to: execute the task script information based on the script parser, and obtain the content returned by the script execution operation.

[0059] It should be noted that the above explanations and descriptions in the embodiment of the reimbursement document review method based on the multimodal large model can also be applied to the reimbursement document review device based on the multimodal large model in the embodiment of the present invention, and will not be repeated here.

[0060] Figure 5An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the computer program in the memory 530 to execute the steps of the reimbursement document review method based on the multimodal large model provided in the above embodiments.

[0061] For example, it includes: obtaining the review request information input by the user regarding the reimbursement document; inputting the review request information into the multimodal big model to obtain the tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and the preset tool information; according to the tool call information sequence, calling the corresponding tools in turn to perform operations and obtain the content returned by each call operation; based on the content returned by each call operation, inputting it into the multimodal big model to obtain the review result of the reimbursement document.

[0062] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0063] On the other hand, an embodiment of the present invention further provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the steps of the reimbursement document review method based on the multimodal big model provided in the above embodiments, including: obtaining the review request information about the reimbursement document input by the user; inputting the review request information into the multimodal big model to obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; according to the tool call information sequence, call the corresponding tools in sequence to perform operations to obtain the content returned by each call operation; based on the content returned by each call operation, input it into the multimodal big model to obtain the review result of the reimbursement document.

[0064] On the other hand, an embodiment of the present invention further provides a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, wherein the computer program is used to enable the processor to execute the steps of the reimbursement document review method based on the multimodal large model provided in the above-mentioned embodiments, including: obtaining review request information about the reimbursement document input by the user; inputting the review request information into the multimodal large model to obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal large model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; according to the tool call information sequence, calling the corresponding tools in turn to perform operations to obtain the content returned by each call operation; inputting the content returned by each call operation into the multimodal large model to obtain the review result of the reimbursement document.

[0065] The processor-readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO), etc.), optical storage (such as CD, DVD, BD, HVD, etc.), and semiconductor storage (such as ROM, EPROM, EEPROM, non-volatile memory (NANDFLASH), solid-state drive (SSD)), etc.

[0066] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0067] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for reviewing reimbursement documents based on a multimodal large model, characterized in that: include: Get the review request information about the reimbursement document entered by the user; Inputting the review request information into the multimodal big model to obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal big model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; According to the tool calling information sequence, the corresponding tools are called in sequence to perform operations, and the contents returned by each calling operation are obtained; The contents returned by each calling operation are input into the multimodal large model to obtain the audit result of the reimbursement document.

2. The method according to claim 1, characterized in that The step of inputting the review request information into the multimodal large model to obtain a tool call information sequence corresponding to the reimbursement document review process includes: Based on the retrieval enhancement generation algorithm, the review request information is matched with the reimbursement document knowledge base to obtain target knowledge information; The review request information and the target knowledge information are input into the multimodal big model to obtain the tool calling information sequence.

3. The method according to claim 1, characterized in that The step of inputting the review request information into the multimodal large model to obtain a tool call information sequence corresponding to the reimbursement document review process includes: Inputting the review request information into the multimodal large model to obtain missing parameter information; Initiating a user dialogue based on the missing parameter information; Receive user reply information, input the user reply information and the review request information into the multimodal big model together, and obtain the tool call information sequence.

4. The method according to claim 1, characterized in that: The tool call information sequence includes API interface call information; calling corresponding tools in sequence to perform operations according to the tool call information sequence, and obtaining the content returned by each call operation, including: Determine the target API interface and target parameter information according to the API interface call information; The target API interface is called based on the target parameter information to obtain the content returned by the target API interface.

5. The method according to claim 1, characterized in that The tool call information sequence includes database query statement information; calling corresponding tools in sequence to perform operations according to the tool call information sequence, and obtaining the content returned by each call operation includes: Execute the database query statement information to obtain the content returned by the database query operation.

6. The method according to claim 1, characterized in that The tool call information sequence includes task script information; calling corresponding tools in sequence to perform operations according to the tool call information sequence, and obtaining the content returned by each call operation includes: The task script information is executed based on the script parser to obtain the content returned by the script execution operation.

7. A reimbursement document review device based on a multimodal large model, characterized in that: include: A first acquisition module is used to acquire the review request information about the reimbursement document input by the user; A second acquisition module is used to input the review request information into the multimodal large model to obtain a tool call information sequence corresponding to the reimbursement document review process; wherein the multimodal large model is used to perform intent recognition based on the review request information, and determine the tool call information sequence based on the intent recognition result and preset tool information; A calling module, used to call corresponding tools in sequence to perform operations according to the tool calling information sequence, and obtain the content returned by each calling operation; The third acquisition module is used to input the content returned by each calling operation into the multimodal large model to obtain the review result of the reimbursement document.

8. An electronic device comprising a processor and a memory storing a computer program, characterized in that: When the processor executes the computer program, the reimbursement document review method based on a multimodal large model described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the reimbursement document review method based on a multimodal large model described in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the reimbursement document review method based on a multimodal large model described in any one of claims 1 to 6 is implemented.

Images